> `follow-mode' has to synchronize the ‘window-end’ of one window with the
> ‘window-start’ of another.
Not when one of them is temporarily obscured by a resized mini-window,
IMO.
> Since mini-window resizing can persistently
> change the end position of the first of these windows
Any command that clears the echo area will resize it back, AFAIK. So
it's not persistent.
> OT1H we do care about point being visible when its window is partially
> obscured by the mini-window and deliberately scroll the window in that
> case. OTOH we'd say that `follow-mode' should not care about keeping
> its text coherent in that case. Is that fair?
Yes. In that situation, the user most probably reads the mini-window
text anyway, and if not, then she reads the text at point. The
probability that she is reading the text that will be scrolled out of
view as result of the mini-window resize is IMO minuscule.